vezz
☆    

Erba (CO), Italy,
2010-03-10 14:00
(5132 d 05:57 ago)

Posting: # 4888
Views: 5,672
 

 Nonparametric confidence intervals [Nonparametrics]

Dear All,

the paper by Willavize and Morgenthien, Nonparametric confidence intervals for Tmax in sequence-stratified crossover studies (Pharmaceutical Statistics 2008), has already been cited in this forum (for example here).

I tried to reproduce the results presented in the paper using the median scaling method (MS) and the simplified nonparametric methd (NP). Both these approaches, as stated in the sections 3 and 4.1, request the point estimates and the confidence limits to be divided by 2 as a final step. Since MS and NP provide estimates of 2*delta, twice the shift in location, this step is necessary to obtain estimates of delta.

I suppose that the results presented in Table II refer to delta. Please note that in this table also the estimates obtained from a simple ANOVA model (indicated as normal theory method, NT) are presented, and clearly their comparison with estimates of 2*delta would make little sense. However, with MS and NP I obtain exactly the same point estimates and confidence intervals reported in Table II skipping the division by 2. In my opinion, all the results presented in the table for MS and NP should be divided by 2 in order to obtain estimates of delta instead of estimates of 2*delta.

Maybe someone else who already tried to apply the methods described in the paper can help me clarify this point? If you are interested, I would be happy to share with you the SAS code used. Thank you very much for your help.

Kind regards,

Stefano
d_labes
★★★

Berlin, Germany,
2010-03-10 16:27
(5132 d 03:29 ago)

@ vezz
Posting: # 4890
Views: 4,714
 

 Willavize CIs

Dear Stefano,

a short check of the median-scaled differences from Table 1 (worked example) of that paper with PROC npar1way and the new HL (Hodges-Lehmann) option in SAS9 proves you right :ok:.

Moreover its obvious that the results of the NP and MS method are consistently higher then the other methods by a factor 2 approximately.

Consider to write to Susan Willavize (Susan.A.Willavize(at)pfizer.com).

Regards,

Detlew
vezz
☆    

Erba (CO), Italy,
2010-03-10 17:24
(5132 d 02:32 ago)

@ d_labes
Posting: # 4891
Views: 4,747
 

 Willavize CIs

Dear D. Labes,

thank you for your prompt reply. I have already tried to contact the author but the e-mail address provided in the paper doesn't work. Anyway, after reading your post I feel reassured that my thinking is correct.

Kind regards,

Stefano
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
91 visitors (0 registered, 91 guests [including 13 identified bots]).
Forum time: 19:57 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5